Thanks to jet lag, I woke up at 1:30 AM this morning, and could not find anything better to do than to read through the 2,631 posts of this amazing thread covering the development of the Soekris dam1021. From it, I gathered quite a few things:
First and foremost, the more I am learning about the R-2R sign magnitude DAM design, the more I like it. As its acronym implies, it’s a modulator, not a converter, and as such, it is a lot more analog than a digital-to-analog converter. Where a DAC works on frequencies in the digital domain, a DAM works on voltages in the analog domain. Both turn a digital signal into an analog one, but they do so in vastly different ways. Getting to appreciate some of these differences took me quite a few posts on the thread.
Second, the Soekris design, much like the converters developed by MSB, implements a sign magnitude ladder, rather than a more traditional R-2R ladder network. As a result, it is much less sensitive to the harmonic distortions introduced by the ladder’s resistors. Nevertheless, you will still want to use resistors with the smallest-possible tolerance, typically 0.01% (can’t get any less), at least for the 14 most-significant bits of the ladder. Fortunately, the 13 less-significant bits and the sign bit can use cheaper resistors with a 0.05% tolerance. This leads to 218 low-tolerance resistors being needed for a pair of mono channels. In the Soekris design, R equals to 4K99, which means that 2R equals to 10K0.
Third, for some reasons that I still fail to fully grasp, the quality of the clock used with an R-2R sign magnitude DAM does not seem to be as important as when used with a delta-sigma DAC. As a result, the Si514 seems to be good enough, even though there seems to be agreement on the fact that an Si570 might deliver slightly better results. And the Soekris design does not support the use of an external clock, even though I’m not clear on whether or not this could be added to the design.
Fourth, everybody seems to be complaining about the output buffers, which are using the LME49724 operational amplifiers. The LME49990 could be used as an alternative, but using raw outputs seems to be the preferred option.
Fifth, the power section should use better capacitors, and so should the output buffers (if used at all). Also, the addition of standalone power regulators might improve things quite a bit, especially for the power section of the shift registers where the analog output is being modulated. Compared to everything else, this part of the Soekris design seems particularly “light” in my opinion.
Last but not least, as one would expect with such a design, the resistor ladder is only half of the story. The other half is found in the FPGA, in the form of filters, and using the wrong one can lead to terrible results. Fortunately, the DIY community seems to be up to the task of developing high-quality filters.
Bottomline, Soekris has done an amazing piece of work, which proves that an R-2R sign magnitude DAM can be developed on a budget. Packing 8 to 10 of them into a small enclosure is another thing altogether, but I am emboldened-enough to give it a try. At this point, the only thing that really worries me is the amount of PCB real-estate required by 3,360 resistors, and the power supply needed to drive all this, even though the whole DAM seems to require only 2.4 W.
Ignorance is bliss…